The support benchmarks worth keeping in the AI era are cost per ticket, one touch resolution and tickets per $100 of revenue. Drop per agent volume quotas, ticket count upgrade triggers and fully human 24/7 staffing. Let AI take routine volume while people own exceptions.
The old staffing question was whether you could cover support around the clock. The better question now is which tickets should never reach a human at all, and which ones should never reach a bot.
A $2 million Shopify brand with two people on support can look fine on paper and still be quietly losing customers. Tickets close, the inbox reaches zero most Fridays, and nobody can say what a ticket costs, how many need three replies, or why December feels like a different job. Then the team adds an AI agent, ticket counts per person drop, and the metrics the manager used to judge performance stop meaning anything.
That is the real shift in ecommerce customer service. AI did not just make support faster. It changed which numbers are worth watching. Some benchmarks built for all human teams still hold, because they measure the business rather than the people. Others now push a team in the wrong direction if you keep them as targets.
I use Gorgias as the helpdesk for Fastlane Media, even though we are a media company rather than a store, so I watch this category from the inbox side as well as from the podcast chair. What follows is the short list of metrics I would keep, the ones I would retire, a worked example on one store, and what I would do this quarter at each stage. If you are doing $10K months, most of this can wait. If you are past $1M, the ticket math in the middle is the part to steal.
The three support metrics that still predict margin once AI answers first are cost per ticket, the share of tickets resolved in one touch, and tickets per $100 of revenue. All three measure the business rather than an individual agent, which is why they survive the move to AI.
Cost per ticket is the one most stores skip. Divide total monthly support cost (wages or outsourcing fees, helpdesk and AI subscriptions, a share of a manager’s time) by tickets resolved. When Gorgias surveyed 131 ecommerce leaders and analyzed more than 780 stores on its helpdesk in 2019, 72% of stores did not know this number, and in my conversations with merchants it is still rarely on the margin sheet. Gorgias’s guide to calculating the marginal cost of support remains a sound method. Once AI is involved, track cost per resolved ticket, not cost per conversation, because a bot that answers and then hands off still cost you the human time.
One touch resolution is the quality signal. That same 2019 Gorgias account data put the average ticket at 1.76 responses to close. Every extra reply is a customer waiting and an agent rereading context. Multi touch tickets cluster in predictable places, usually returns, exchanges and delivery exceptions, so this metric tells you where to fix the process rather than where to add people.
Tickets per $100 of revenue is the planning number. The same dataset showed roughly 1.1 tickets per $100 of revenue for smaller stores and 0.4 for larger ones. Treat those as dated reference points from one helpdesk’s customers, not a universal norm. The value is in your own trend: a rising ratio usually means a prevention problem in product pages, shipping updates or the return flow, while a rising cost per ticket usually means the tickets left for people are getting harder.
Preventing a ticket is worth more than answering it faster, because the cheapest ticket is the one a customer never needs to send. Most “where is my order” and “what’s my tracking” volume comes from silence after checkout, not from anything an agent can fix in the reply.
Proactive order updates are the first lever. Branded tracking pages, delay alerts and shipping texts cut that volume before it reaches the inbox. Tools like IMpower pipe text replies straight into the helpdesk, so an SMS update can become a conversation without a new channel to monitor. Peak season makes this obvious, which is why fixing BFCM shipping and tracking before Black Friday is a support decision as much as a logistics one.
The second lever is the boring list: product pages that answer sizing and compatibility, a maintained FAQ, a short onboarding message after purchase for anything that needs setup, a clear checkout and honest delivery estimates. Fulfillment partners such as ShipBob matter here too, because faster and more predictable shipping removes a whole category of contact. And returns deserve their own audit. A return policy that is prompt and fair produces fewer second and third replies than any macro an agent can write.
On site chat is now the default channel for most Shopify brands past their first year, and the debate has moved. It used to be about cost, because someone had to be online in real time. An AI agent answering from your policies and order data removes most of that cost for routine questions, so the real decision is handoff rules: what the agent answers, what it passes to a person, and how quickly that person picks up.
Four support targets should be retired once AI handles first line questions: per agent ticket quotas, ticket count as a trigger for platform upgrades, fully human 24/7 staffing, and third party chat widgets bolted onto product pages. Each made sense for an all human team, and each now points a team in the wrong direction.
The per agent quota is the one I would drop first. Volume targets in the region of a thousand tickets per agent a month were built on easy tickets: order status, policy lookups, password resets. Those are exactly the tickets an AI agent should absorb. A person measured on raw volume after that shift is either rushing escalations to hit the number or being judged against work that no longer exists. Measure resolution quality, time to escalate and repeat contact instead.
Ticket count as an upgrade signal was always shaky. Ticket volume is a side effect of order volume, not a reason to change plans. The decision to move to Shopify Plus rests on revenue, checkout and operational needs, and your support load should not be the trigger.
Fully human 24/7 coverage solved a real problem the wrong way. Stores never close, and customers shopping at 2am still expect answers. The only option for an all human team was outsourced overnight shifts. For most brands the realistic answer now is AI coverage for routine questions around the clock, with a clear, stated promise about when a person picks up everything else.
Finally, chat widgets from social platforms embedded on product pages mostly faded. Owning the conversation on your own site, inside your own helpdesk, keeps the order context and the customer history in one place. If an old widget is still sitting on your product pages, take it off and route that traffic to your main chat.
Running the ticket ratio on a single store turns a vague “we need help with support” into a staffing and tooling decision you can argue about. Here is how I would work it for a mid sized brand that has not added AI yet.
The store below is a labelled composite, built from patterns I see in merchant conversations and on the podcast, not a real brand. Every figure is illustrative. Assume a Shopify apparel brand doing $2 million a year, two people answering support, a helpdesk already in place, and no AI agent.
Illustrative example, assuming the 1.1 and 0.4 tickets per $100 reference points roughly apply. At 1.1 tickets per $100, $2 million in revenue implies about 22,000 tickets a year, or around 1,830 a month. At 0.4, it implies about 8,000 a year, or roughly 670 a month. A $2 million brand usually sits between the two, so call it 1,000 to 1,500 tickets a month. Two people answering 1,250 tickets a month while also handling returns, reviews and social comments are stretched, and December, when volume and response times both spike, breaks them.
Here is what I would do with that store. Run the cost per ticket calculation this month, before any hiring decision. Fix the return flow first, because that is where multi touch tickets pile up and where one touch resolution improves fastest. Then change the staffing question: before hiring a third person, run an AI agent on order status, sizing and return policy questions for 90 days and measure cost per resolved ticket and satisfaction on AI handled conversations. Drop any per agent volume target entirely. And remove one thing: if a leftover social chat widget sits on product pages, take it off and route that traffic to the main chat.
You can disagree with the order. A brand with a strong founder voice might add AI later and protect the personal tone longer. That is a fair call under about $1 million. At $2 million and two burned out agents, I would not wait.
Phone support is worth reconsidering because AI voice agents have cut the cost of answering a call so far that the old objection, that phone is too expensive to staff, no longer settles the question. A number that only goes to voicemail is still not phone support, but a voice agent that answers and hands off can be.
On episode 463 of the eCommerce Fastlane podcast, I put the cost shift in plain numbers: a human answered call used to run around $7 to $12, and an AI voice agent call is about 40 cents. Those were the figures I cited on the show, not an audited benchmark, so treat them as a direction rather than a promise. The reason a voice agent can lift satisfaction instead of hurting it is simple, as I said on that episode: “the agent picks up on the first ring, it’s available whenever the customer wants to communicate with the brand.”
The same episode was clear about stage: “So for starting merchants, probably this is overkill for you.” The exception is a founder drowning in order status questions, and even then the first fix is usually inbox speed and better shipping updates. For scaling merchants, a 90 day pilot that measures cost per resolved call and satisfaction is the honest test. Tools like Aircall still matter for brands that want human phone teams with order context beside the call, because a phone line that cannot see the customer’s orders recreates the multi touch problem by voice.
One boundary I hold for voice and chat alike: “I would keep the human in the loop on the final transaction, to be quite honest with you.” An agent that answers, explains and recommends is useful. One that completes purchases or approves exceptions without a person is where brands get burned.
AI agents fit best on top of the helpdesk you already run, answering repeatable questions from approved store data, while people own refunds, exceptions, angry customers and anything that needs judgment. That split is what makes agent productivity jump: not humans working faster, but humans no longer touching the easy tickets.
Analysts expect the shift to keep going. Gartner predicted in March 2025 that agentic AI will autonomously resolve 80% of common customer service issues by 2029, with a 30% reduction in operational costs. That is a forecast, not a measurement, and it covers all industries, not Shopify stores in particular.
Customers are less enthusiastic. A Gartner survey of 5,728 customers in December 2023 found 64% would prefer companies did not use AI for customer service, and 53% would consider switching to a competitor if they learned a company was going to. The top concern was difficulty reaching a person. That is the real design constraint: the bot has to make reaching a human easier, not harder.
The architecture point came through clearly when I talked with Ashu Dubey on episode 327, when his company was still called Gleen AI (it has since rebranded as Alhena AI). His rule was that an AI layer should slot into the tools a team already uses: “if they are using a help desk, [Gleen] has to work on top of that help desk, not replace that help desk.” Whether you use the AI built into Gorgias or a separate agent, the setup work is the same, and our guide to setting up an AI chatbot on Shopify without losing customer trust covers the boundaries to set before switching one on.
The right move this quarter depends on your volume: under about $500K, fix prevention and speed in one inbox; between $500K and $5M, measure cost per ticket and pilot one AI channel; above that, rebuild your support metrics around resolution rather than agent throughput. As I said on episode 463, “don’t try all of these things all at once because really you should just pick one and pick one that really matches your current bottleneck literally in your business like today.”
If you are under $500K, the founder answering is still your best service, and prevention is your roadmap: clearer product pages, a real FAQ, shipping updates and a simple return flow. Our piece on personalized customer service examples that use context, not just a first name shows why that founder touch is worth protecting.
Between $500K and $5M, work out your cost per ticket this month, then compare your ticket ratio to the 0.4 and 1.1 reference points. If you are close to 1.1, the problem is prevention, not staffing. Once AI takes first line tickets, stop measuring people on volume. Set a kill rule for any AI pilot before it starts: if satisfaction on AI handled conversations drops or customers complain they cannot reach a person, pull it back to a narrower set of questions. Before peak season, decide which questions an AI agent answers and which go to people, the split covered in our guide to customer service automation during a sales spike.
Above $5M, retire volume quotas, set escalation time targets for human tickets, and audit where customers struggle to reach a person. Support and retention are the same conversation at this size, which is why loyalty platforms like Smile.io sit so close to the helpdesk in a mature stack. The brands that keep loyal customers are the ones whose support feels faster with AI and still human when it counts.
A good first response time is near instant for routine questions answered by chat or an AI agent, and within a few business hours for anything a person must handle. Gorgias’s 2019 helpdesk data showed an average of 5 hours 13 minutes, rising to more than 10 hours in December, a useful reminder of how slow all human queues get at peak. Customers now expect order status, return policy and sizing answers immediately, because AI agents can supply them. What they will still accept is a short wait for a human on a genuine problem, as long as you tell them when someone will reply. Set two targets rather than one: instant for routine, and a stated promise for escalations.
There is no useful universal number of tickets per agent anymore, because AI absorbs most of the simple tickets that inflated volume benchmarks. Older quotas in the region of a thousand tickets per agent a month assumed people handled order status and policy questions. Once an AI agent takes those, the tickets left for people are slower and harder, so a volume quota pushes agents to rush the cases that most need care. Track resolution quality, time to escalate, repeat contacts and customer satisfaction on human handled tickets instead. Use volume only to spot staffing gaps, never as a personal target.
To calculate the cost of a support ticket, divide your total monthly support cost by the number of tickets resolved that month. Total cost includes agent wages or outsourcing fees, helpdesk and AI tool subscriptions, and a share of the manager’s time. In a 2019 Gorgias survey, 72% of ecommerce stores did not know this number, even though it belongs in margin planning next to average order value and acquisition cost. Once you have it, compare it with your ticket ratio per $100 of revenue. A rising cost per ticket usually means complex tickets are growing, while a rising ratio usually means a prevention problem in product pages, shipping updates or returns.
A small Shopify store should usually wait on a full AI agent and fix prevention and response speed first. Under about $500K a year, the founder answering personally is often the brand’s best service, and most tickets come from fixable gaps such as unclear shipping times or a confusing return flow. A Gartner survey found 64% of customers would prefer companies did not use AI for customer service, so a bot that blocks access to a person costs trust a small brand cannot spare. The exception is a store drowning in order status questions. Then a narrow AI setup that answers only from your policies and tracking data, and hands everything else to you, can buy back hours.
Phone support is worth adding for scaling brands with high order value or complex products, mainly through AI voice agents that answer on the first ring and hand off to people. AI voice has cut the cost of answering a call sharply, which removes the old objection that phone was too expensive to staff. For a starting merchant it is usually overkill, and a voicemail box that turns into a ticket is still not phone support. If you are between roughly $500K and $5M, run a 90 day pilot and measure cost per resolved call and satisfaction before committing. Keep a person in the loop for refunds, exceptions and any purchase decision.
